[Date Prev][Date Next][Thread Prev][Thread Next][Date index][Thread index]
RE: st: RE: Panel Data Problem
To add to my earlier comments, and
as discussed just now in another
thread, I don't see interpolation
as necessarily amenable to general-purpose
> It's both an answer and a question.
> It's a question because I was surprised Dr Cox did not
> include -mvis- as
> an option. It's an answer because multiple imputation is a
> common way of
> dealing with missing values. Obviously applying it to a panel
> data with so
> many missing values will itself be a statistical challenge.
> Is that a question or an answer?
> I'll in turn give an answer that is
> really a question.
> As one gets into more and more elaborate imputation
> of missing data, what then is the status
> of any model later fitted to the data?
> I am reminded of those stories of naive
> students rediscovering the very relationships
> used to fill in the gaps in scrappy data by
> economists working for international organisations.
> > What about -mvis- for multiple imputation?
> > You have a lot of missing data here.
> > If -ipolate- helps little, it seems
> > unlikely that any other method will
> > help more.
> > You might get marginal improvements
> > by interpolating in log price, then
> > exponentiating. That is, a zeroth
> > approximation in most economies
> > would be that prices are increasing
> > multiplicatively.
> > > I have a panel data of daily price of a commodity in
> > > different locations of a city over 7 -year period.
> > > Unfortunately about 40 % data is missing from both
> > > right and left (dependent variable) side of the model.
> > > How should I go about it? I tried Stata's ipolate
> > > command to fill up the missing observations but did
> > > not help much. When I ran xtreg with random effect
> > > option (re) a number of records are dropped. I want to
> > > capture as many panels as I can. Any help is greatly
> > > appreciated.
* For searches and help try: